Artists score key court win against Midjourney and Stability AI

Midjourney once compiled a style cheat sheet listing 4,700 artists so users could mimic their work with a prompt. A federal judge just decided that list might actually cross the line into trademark infringement.

U.S. District Judge William Orrick advanced key copyright and trademark infringement claims against Stability AI, Midjourney, DeviantArt, and Runway. It marks a pivotal procedural victory for visual artists who argue their work was scraped without consent to train AI image generators.

While the judge dismissed several secondary claims around contract breach and digital copyright management, the core case survived. That means the lawsuit is officially moving to discovery, where artists can finally peek behind the industry's closely guarded curtain.

A prompt sheet turns into legal liability

The amended complaint had far more teeth than its earlier version, which Judge Orrick had trimmed back last year for lack of specifics.

This time, the artists homed in on the infamous Midjourney Style List. The feature let users generate images matching the signature aesthetic of 4,700 real-world artists, without their knowledge or permission. The judge agreed that promoting this list could mislead users into believing those creators endorsed the platform, allowing the trademark claim to proceed.

Stability AI took an even heavier hit. The judge found sufficient grounds that Stable Diffusion was largely built on copyrighted works and, crucially, may have been created to facilitate infringement by design. That allowed a claim of induced copyright infringement to move forward alongside direct infringement.

The defense had tried to argue that each artist needed to identify every single image used in the LAION dataset of 5 billion scraped pictures. Judge Orrick rejected that hurdle entirely, citing the sheer scale of the data and the deliberate opacity of the software.

Downstream tools are no longer safe

Perhaps the most alarming part of the ruling for the broader software ecosystem is that DeviantArt is still on the hook.

DeviantArt didn't train Stable Diffusion from scratch; it simply packaged the open-source weights into its own DreamUp image tool. Its lawyers previously warned the court that letting the case advance would unleash havoc on an innumerable number of companies that simply deploy open-source models.

The judge advanced the claims anyway, pulling downstream integrators right into the crosshairs alongside Runway, the startup that helped build the early versions of the model.

Following the ruling, plaintiff Kelly McKernan called it a huge win on X, writing that artists now get to uncover all the things those companies don't want the public to know.

Why it matters

This ruling fundamentally shifts the balance of power in AI copyright battles by taking away two favorite defense arguments: the black box defense and the wrapper defense.

For visual artists and creators, reaching discovery is the real jackpot. Even if this case eventually settles before a final verdict, getting access to internal emails, scraping scripts, and training run logs will permanently expose how early generative models processed creative work.

For developers and startups, the inclusion of DeviantArt is a loud wake-up call. The court is signaling that wrapping or fine-tuning an open-source model does not shield you from liability if the base model itself is built to facilitate infringement. Relying on an upstream provider to take the legal bullet looks like a failing strategy.

While tech giants like OpenAI and Google can afford to sign multimillion-dollar licensing deals with media publishers to protect their flanks, smaller players like Stability AI and Midjourney have far less cash on hand. If training on scraped datasets loses its fair use shield, their foundational business models will be on the line.